---
license: cc-by-4.0
datasets:
- APRIL-AIGC/UltraVideo
base_model:
- Wan-AI/Wan2.1-T2V-1.3B
---

# UltraWan LoRA — ComfyUI Conversion (Unofficial)

**This lora only works with the T2V 1.3b model, it does not work with the 14b. The 4k model has some problems, but the 1k model works well.**

This repository contains **converted LoRA weights** from the official [**APRIL-AIGC/UltraWan**](https://huggingface.co/APRIL-AIGC/UltraWan) release.  
The only modification made was converting the `.ckpt` / original format into `.safetensors` compatible with **ComfyUI**.  
No training, fine-tuning, or architecture changes were performed.

---

## 📌 What This Is

- ✅ Direct conversion of original UltraWan LoRA weights to **ComfyUI-compatible safetensors format**
- ❌ Not trained, modified, or altered in any way beyond the format change
- 🎯 For users who want to test **UltraWan LoRA** with ComfyUI’s Text-to-Video pipelines

---

## 🔗 Attribution

- **Original Model**: [**APRIL-AIGC/UltraWan**](https://huggingface.co/APRIL-AIGC/UltraWan)
- **Base Model Architecture**: [TencentARC/Wan](https://github.com/TencentARC/Wan)
- **License**: CC-BY-4.0 with Additional Restrictions (see below)

> I do **not** claim ownership of the original model, dataset, or LoRA weights.  
> All credit goes to the **APRIL-AIGC** team and the authors of the [UltraVideo Dataset](https://arxiv.org/abs/2506.13691).

---

## 📄 License & Use

These weights are subject to the license terms of the **UltraVideo Dataset**, which include:

- ❗ **Non-commercial research only**
- ⚠️ No redistribution of raw videos or identifiable content
- ✅ You may use these weights to run inference in ComfyUI for research/academic purposes
- ❌ You may **not** use these weights or outputs for commercial purposes, product development, or monetization

Please see the full license in [`LICENSE`](./LICENSE).

---

## 🔧 Recommended LoRA Strength

| LoRA Version       | Recommended Strength |
|--------------------|----------------------|
| `ultrawan_1k`      | **0.25**             |
| `ultrawan_4k`      | **0.5**              |


![image/png](https://cdn-uploads.huggingface.co/production/uploads/6471da9222016353ae38b6bc/GR6raXh4OSBgpBHtOq7ml.png)

> You may experiment with different values, but these are the recommended starting points for best results.

---

## 📂 Contents

| File | Description |
|------|-------------|
| `ultrawan_1k_comfy.safetensors` | LoRA converted to safetensors, ready for ComfyUI |
| `ultrawan_4k_comfy.safetensors` | LoRA converted to safetensors, ready for ComfyUI |
| `LICENSE` | UltraVideo Dataset License (CC-BY-4.0 + Restrictions) |
| `README.md` | This file |

---

## 🧩 How to Use in ComfyUI

Place the `.safetensors` file in:

```
ComfyUI/models/loras/
```

Then use a **LoRA Loader** node in your workflow and select the LoRA file from the dropdown menu.

---

*This is a community-contributed format conversion. All rights belong to the original creators.*